GRTR: Drug-Disease Association Prediction Based on Graph Regularized Transductive Regression on Heterogeneous Network

Abstract

Computational drug repositioning helps to decipher the complex relations among drugs, targets, and diseases at a system level. However, most existing computational methods are biased towards known drugs-disease associations already verified by biological experiments. It is difficult to achieve excellent performance with sparse known drug-disease associations. In this article, we present a graph regularized transductive regression method (GRTR) to predict novel drug-disease associations. The proposed method first constructs a heterogeneous graph consisting of three interlinked sub-graphs including drugs, diseases and targets from multiple sources and adopts preliminary estimation of drug-related disease to initial unknown drug-disease associations for unlabeled drugs. Since the known drug-disease associations are sparse, graph regularized transductive regression is used to score and rank drug-disease associations iteratively. In the computational experiments, the proposed method achieves better performance than others in terms of AUC and AUPR. Moreover, the varying of parameters is shown to verify the importance of preliminary estimation in GRTR. Case studies on several selected drugs further confirm the practicality of our method in discovering potential indications for drugs.